Hospitals & AI: Innovative MedTech Advances Healthcare Treatment

The goal behind a smart hospital is to improve several key factors on significant healthcare inflection points, including clinical outcomes, operational efficiency, patient experience in both guidance and engagement and safety, and data usage to inform decision-making. However, none of these desired improvements would be possible unless the hospital’s various internal silos are dismantled first and then integrated with the hospital’s wider ecosystem.

And just like the smart city, which aims to incorporate advanced technologies throughout a city’s infrastructure to help improve safety, mobility, sustainability, and economic vitality for residents, the trend in smart hospitals is one that intends to benefit all stakeholders. The smart hospital has the potential to introduce efficiency in times of crisis by enabling staff to focus on the most critical patients. This can be impactful on many levels, including answering patient’s questions on-demand, enabling nurses to devote more time to urgent care, and increasing staff and patient safety by limiting exposure throughout the treatment process.

Most hospitals have yet to achieve “smart” status, and few solutions can be considered complete or available at the point of care. But many are well on the way, having optimized new interconnected assets, clinical processes, and management systems to provide a level of insight not previously possible.

A wide variety of technologies can be brought together to improve a hospital’s capability to gain the insight it needs on a patient population’s condition and the hospital’s own operations. As a result, trying to define what precise combination of technology makes for a smart hospital is oftentimes a moving target, dependent on a varied mix of software, analytics, connected imaging modalities, cloud services, smart building technologies, and patient engagement platforms. However, one succinct characterization of a smart hospital is that it is enabled by a combination of hospital information systems, IoT equipment, AI computing, and sensor fusion.

Two key companies driving and enabling innovation in healthcare are Intel, with its Intel Core i7 processor and Intel Active Management Technology; and NVIDIA, with its CLARA Guardian and EGX Edge AI Stack. Both companies provide platforms that actively facilitate AI in the cloud and at the edge. Sensors, according to Kimberly Powell, Vice President of Healthcare at NVIDIA, supply the eyes, ears, and voice in the hospital, culminating into robots and computer vision. Meanwhile, edge computing enables the real-time, secure deployment of applications—given that AI processing and sensor fusion can automate insights and actions. For their part, cameras and computer vision can enable the real-time monitoring of patient status and operational events, along with body temperature and fall detection. Pairing computer vision with the mobility of a robot is tantamount, then, to gaining a new member to the care team. More detail on autonomous robot development is discussed later in this insight.

“Clara Guardian powered end-to-end AI healthcare solutions are now deployed across 130+ healthcare facilities, helping hospitals saved millions of dollars in operational expenses, and improving patient experience with automation. The future of healthcare will be powered by AI.”

NVIDIA, the highly regarded maker of trail-blazing graphics processors and accelerated computing chips, is harnessing its considerable expertise in machine learning and artificial intelligence (AI) to bolster healthcare. During the company’s annual GPU Technology Conference (GTC), NVIDIA sought to foster discussion on subjects such as high-performance computing, graphics, and healthcare. The conference featured more than 120 sessions related to healthcare alone, such as accelerating drug discovery and genomic analysis, designing next-generation hospitals, patient monitoring, and training AI for medical imaging. The focus today on healthcare by NVIDIA is understandable. “The world is confronting COVID-19, one of the greatest challenges in human history,” said NVIDIA CEO Jensen Huang in remarks during the company’s annual meeting of stockholders.

To that end, NVIDIA’s newest AI supercomputer and NVIDIA-accelerated computing are being deployed in the scientific community to help sequence and image the novel coronavirus causing COVID-19, to search for vaccines and treatment, and to build robots. Innovation has been ongoing for some time regarding surgical robotics, but other logistical areas of innovation include disinfection and assistive robotics.

Robots can take various forms.
Telepresence robots, for instance, feature remote control capabilities with bidirectional audio and video. Companies making telepresence robots include Silicon Valley’s Anybots, Double Robotics, Suitable Technologies, and OhmniLabs; Massachusetts-based Ava Robotics; VGo Communications in New Hampshire; and Axyn Robotics from France.

Another type, the autonomous UV disinfection robot, ensures areas are safely and properly disinfected without the need for human intervention, keeping people away from dangerous and toxic chemicals. The makers of autonomous UV disinfection robots include Altoros in Silicon Valley, Denmark’s Blue Ocean Robotics, and Sterilray in New Hampshire.

Then there are more advanced robots in development that can manipulate their environment. Examples include TRINA, currently under development at the Intelligent Motion Laboratory at the University of Illinois at Urbana-Champaign; and Moxi from Diligent Robotics.

TRINA, or Tele-Robotic Intelligent Nursing Assistant, is a mobile manipulation robot with telepresence capabilities that is designed to let medical staff perform a variety of routine tasks, such as bringing food and medicine, moving equipment, cleaning, and monitoring vital signs, while communicating with the patient. The latest iteration, TRINA 2.0, has a slim profile for navigating tight spaces along with more precise manipulation capabilities for handling small items.

Researchers are also testing the best technique for manipulating the robot, such as direct teleoperation and supervisory control. Direct teleoperation provides an immersive experience, while supervisory control allows the operator to oversee a semi-autonomous robot. The caveats include what users say is a user interface experience that feels strange, along with a speed during operation that is much slower compared to that of a real nurse.

Moxi differs from TRINA in that it is smaller, has a single arm, and is completely autonomous. The overall design is intended to be more socially engaging as well, as Moxi is intended to help clinical staff with non-patient-facing tasks, such as gathering items from supply closets and bringing them to patient rooms, delivering lab samples, and removing soiled linen bags. The CEO of Diligent Robotics, Andrea Thomaz, stated that Moxi’s mission is focused on relieving nurses’ tasks by giving them more time to concentrate on patients. With the outbreak of COVID-19, such a mission has a newfound purpose and urgency, Thomaz said.

The type of automation being built for Moxi may help hospitals maintain consistent care workflows and give staff more time to care for patients. The company chose to develop Moxi in hospitals as their environments provide a semi-structured setting that requires efficiency and consistency of care. Moxi could also help free up time for nurses, who spend up to 30% of their day on non-value-added tasks. During the beta trial, Moxi was deployed for over 120 days in four hospitals, working alongside more than 125 nurses and clinicians. To operate in busy hospital environments, Moxi’s AI framework was continually training features for social awareness, mobile manipulation, and direct human-guided learning—all of which is needed by hospitals to adapt to changing workflows.

AI to support healthcare research
NVIDIA recently electrified the technology universe when it announced it would acquire venerable UK chip designer ARM from Japan’s SoftBank Group for $40 billion. By combining NVIDIA’s deep and substantive knowhow in AI with ARM’s extensive ecosystem, NVIDIA hopes to bring the power of AI and

high-performance computing to practically every smart or IoT-connected entity in the world—from smart meters to smart cars, and everything else in between. NVIDIA will continue ARM’s open-licensing model and customer neutrality, and with NVIDIA technology will expand ARM’s IP licensing portfolio. As part of NVIDIA, ARM will continue to operate its open-licensing model while maintaining global customer neutrality, a practice that has been foundational to ARM’s success, with 180 billion chips shipped to date by its licensees. ARM partners will also benefit from both companies’ offerings, including Nvidia’s numerous innovations.

NVIDIA is building the most powerful computer in the UK, -Cambridge-1. The computer will utilize AI to support healthcare research and serve as a hub of innovation to further work being done in critical healthcare and drug discovery. The NVIDIA technologies profiled below are key to understanding the company’s approach to healthcare.

NVIDIA BlueField-2 DPU and accelerated performance
A new class of processor that is software programmable, DPUs—or data processing units—are multi-core CPU systems- on-chips combining a high-performance network interface and a set of programmable acceleration engines that improve the performance of AI applications. At GTC, NVIDIA announced the company’s BlueField-2 DPU, the first data-center-infrastructure-on-a-chip architecture (DOCA) designed to optimize enterprise data centers. The aim is to deliver a range of accelerated software-defined networking, storage, security, and management services running on BlueField DPUs. A single BlueField-2 DPU delivers services that would require as many as 125 CPU cores.

NVIDIA also announced BlueField-2X, the first AI-Powered DPU, combining the features of BlueField-2 with NVIDIA Ampere GPU technology to enable real-time security analytics. Moreover, the NVIDIA EGX AI platform is expanding to combine the NVIDIA Ampere architecture GPU and BlueField-2 DPU on a single PCIe card, providing enterprises with a common platform to build accelerated data centers.

An example provided of the NVIDIA EGX AI platform in use for healthcare was the Northwestern Memorial Hospital in Illinois, which is working with the London, UK-based startup Artisight. Artisight and its IoT sensor platform. Artisight, a deep learning technology provider for the healthcare industry, is using a network of thermal cameras located at hospital entrances to screen people for fever. The approach has reduced the staff needed to monitor the entrances and has also shortened wait times for patients. The startup then provided remote viewing of COVID-19 rooms to limit exposure of the nurses to the novel coronavirus and to decrease the need for personal protective equipment. This was enabled with cameras that feature night vision and microphones using Artisight’s network of NVIDIA GPUs to transcode the video streams for secure viewing on any hospital display.

NVIDIA Clara Discovery and computational drug design
Clara Discovery is a collection of frameworks, applications, and AI models enabling GPU-accelerated drug discovery, with support for research in genomics, proteomics, microscopy, virtual screening, computational chemistry, visualization, clinical imaging, and natural language processing (NLP).

One company utilizing NVIDIA GPUs to power drug discovery is New York-based Schrodinger, an outfit that developed a physics-based computational platform now being used by top biopharma players, many of which have standardized on the company’s platform. The computational approach allows Schrodinger to calculate as many as 3,000 possible compounds and petabytes of data for every drug candidate being considered—an approach that not only delivers faster results but also tackles a scope of analysis far greater than any human could possibly undertake.

And working with an alliance of biopharma companies including Takeda of Japan, Novartis of Switzerland, US outfit Gilead Sciences, and Wuxi AppTec of China, Schrodinger is collaborating with the group in the search for novel antiviral therapeutics to battle COVID-19.

For more information on computer modeling in drug discovery, see Vamstar’s insight titled: In silico & AI: Computer simulation in drug discovery.

NVIDIA DGX SuperPOD and AI supercomputing
The NVIDIA DGX SuperPOD is a unique AI supercomputing infrastructure that deploys as a fully integrated system in weeks, instead of the typical six-month time frame. Cambridge-1, slated to become the UK’s most powerful computer, is intended for healthcare industry research. Expected to be operational by the end of this year, NVIDIA is working with UK-based outfits GlaxoSmithKline, AstraZeneca, -Guy’s and St Thomas’ NHS Foundation Trust, King’s College London, and Oxford Nanopore to use the Cambridge-1.

Part of the DGX SuperPOD configuration is Mellanox CS7500 and CS7520 InfiniBand switches. NVIDIA acquired Mellanox Technologies earlier this year for $7 billion, bolstering the company’s offering for high-bandwidth connections in next-gen data centers.

NVIDIA Jarvis and conversational AI
NVIDIA Jarvis is an application framework for multimodal conversational AI services that delivers real-time performance on GPUs. Jarvis could be an interface for telemedicine and appointment making, and potentially a patient journey guide for many more touchpoints between patients and care providers.

The market and other players
Vamstar estimates that the market for semiconductors supporting healthcare-related AI passed the $1 billion mark in 2020 and will likely quadruple over the next four to five years. In addition to Nvidia, several leading chip vendors with a focus on AI for data centers include Amazon Web Services, AMD, Google, Huawei, Intel, Qualcomm, and Xilinx.

• Amazon Web Services recently released AWS Inferentia, Amazon’s first chip designed for high-performance inference in the cloud. The chip stems from Amazon’s 2015 acquisition of Israeli-based Annapurna Labs.
• AMD has been working to stay competitive in the AI space with AMD EPYC and Radeon Instinct processors.
• Google started using its Tensor Processing Unit AI accelerator ASICs internally in 2015, and in 2018 made them available for third-party use. They are designed specifically for neural network machine learning, using Google’s own TensorFlow software.
• Huawei launched its own AI processor, the Ascend 910, during late 2019 along with the company’s own AI computing framework, MindSpore.
• Intel’s AI offering is its Xeon Scalable Processors. Intel recently ended development on the neural network processors of its 2016 acquisition Nervana Systems to focus on the AI chip architecture obtained with the acquisition of Habana Labs Ltd. Both Nervana and Habana chips are designed for the same purpose, but the Habana design is more powerful and has been shipping since 2018.
• Seeking growth beyond the slowing smartphone market, Qualcomm is introducing the company’s first discrete dedicated AI processor, the Cloud AI 100, designed for data center inferencing work. In September 2020, Qualcomm stated that the Cloud AI 100 is currently being sampled by multiple customers.
• Xilinx, a leader in programmable logic devices such as FPGAs, has developed its own Vitis AI development platform for use with Xilinx FPGAs and the Xilinx adaptive compute acceleration platform (ACAP). FPGAs have an advantage over ASICs as they can be reconfigured in the field, which is important when underlying AI technology is changing rapidly.